Glossary — Language models

Large language model

A large language model is a program trained on an enormous amount of text to predict which token comes next.

3 tokensLargelanguagemodel

Simply put#

Everything ChatGPT, Claude and Gemini do comes from one ability: predict the next token. It is done again and again, token by token, until the answer is finished.

That sounds far too simple to work. But once a model has seen enough text, prediction starts to look like understanding.

Example#

Type "The capital of Estonia is" and the most likely next token is "Tallinn". The model does not "know" anything about Tallinn — it has simply seen that pattern millions of times.

Why it matters to you#

Everything else follows from this: why AI sounds confident even when it is wrong, and how it can write things nobody has ever written.

Related terms